Closed
添加moe_init_routing_v4和moe_re_routing_v2算子的st&标杆 #4543
cpy_123456Add on Aug 24Closed at Aug 25
weihao18
Aug 24 Comments:
Aug 24 Comments:
/assign @cpy_123456


Aug 24 Assign cpy_123456 as responsible person
Aug 25 Close issue
Aug 25 Added label: resolved
Sep 3 Added label: requirement
Sep 3 Removed label: requirement
Sep 3 Added label: torch-npu
Sep 3 Removed label: torch-npu
MoE 系列算子 ST 用例描述与执行报告
一、moe_init_routing_v4(25 个用例,全部 PASS)
1.1 场景覆盖矩阵
int8float16bfloat16float32-1(非量化,DropLess)1(动态量化)1(直方图统计)2(key-value 对)0(DropLess 模式)True0(gather)1(scatter)1.2 用例分组描述
第 1 组:标准场景(small shape, k 较小)
用例编号:
[0][2][4][6][8][10][12][16][18][20][22][24]第 2 组:大 k 场景(fuzz, k=270~2930)
用例编号:
[1][3][5][7][9][11][13][15][17][19][21][23][15](59296×6553)和[3][17](69513×3719)第 3 组:row_idx_type=0(gather 模式)
用例编号:
[4][5][7][8][11][14][20][21][24]1.3 用例明细
1.4 执行时间统计
moe_init_routing_v4_0002— 7.60 usMoeInitRoutingV4_bfloat16_ND_fuzz_000092— 157.09 us二、moe_re_routing_v2(25 个用例,全部 PASS)
2.1 场景覆盖矩阵
int8float16bfloat160(正向索引,permute_token_idx=src_offset)1(反向索引,permute_token_idx=dst_offset)2.2 用例分组描述
第 1 组:单 rank 单 expert(基础路由)
用例编号:
[0][2][4][6][8][9][10][14][16][19][22][24]第 2 组:多 rank 多 expert(EP 场景)
用例编号:
[5][11][12][15][17][20][21][23]第 3 组:大 hidden_size(带宽压力)
用例编号:
[0][3][12][13][14][16]第 4 组:idx_type=1(反向索引)
用例编号:
[2][3][6][7][10][11][14][15][18][19][22][23][24]2.3 用例明细
2.4 执行时间统计
MoeReRoutingV2_int8_ND_fuzz_000300— 3.77 usMoeReRoutingV2_bfloat16_ND_fuzz_000237— 6.65 us三、总体汇总
所有用例均从源 CSV(
/workspace/log/ttk/st/moe_init_routing_v4.csv、/workspace/log/ttk/st/moe_re_routing_v2.csv)中提取带topk_weight的用例,覆盖了 4 种数据类型(int8/float16/bfloat16/float32)、2 种量化模式、2 种索引模式、有/无 scale/scales 输入、不同 rank/expert 组合等关键场景。四、测试命令参考
# 环境准备 source /home/developer/Ascend/cann/set_env.sh export LD_LIBRARY_PATH=/home/developer/Ascend/cann-9.1.0/opp/vendors/custom_transformer/op_api/lib:${LD_LIBRARY_PATH} cd /workspace/ops-test-kit # moe_init_routing_v4 kernel 测试 python3 -m ttk kernel \ -i /workspace/ops-transformer_3876/moe/moe_init_routing_v4/tests/st/test_moe_init_routing_v4.csv \ -o /tmp/result_init_routing_v4.csv \ -d false -b release --no-memory-check \ --plugin /workspace/ops-transformer_3876/moe/moe_init_routing_v4/tests/assets/golden.py \ --pc 1 --ti=0-24 # moe_re_routing_v2 kernel 测试 python3 -m ttk kernel \ -i /workspace/ops-transformer_3876/moe/moe_re_routing_v2/tests/st/test_moe_re_routing_v2.csv \ -o /tmp/result_re_routing_v2.csv \ -d false -b release --no-memory-check \ --plugin /workspace/ops-transformer_3876/moe/moe_re_routing_v2/tests/assets/golden.py \ --pc 1 --ti=0-24五、测试文件清单
moe_init_routing_v4
tests/assets/golden.pyMoeInitRoutingV4TestSpeckernel golden +AclnnMoeInitRoutingV4TestSpecaclnn golden)tests/st/test_moe_init_routing_v4.csvtests/st/test_aclnn_moe_init_routing_v4.csvmoe_re_routing_v2
tests/assets/golden.pyMoeReRoutingV2TestSpeckernel golden)tests/st/test_moe_re_routing_v2.csv